Calva Backseat Driver MCP Server

Overview

The Calva Backseat Driver is a purpose‑built MCP server that turns any AI assistant—such as CoPilot or Claude—into an interactive Clojure programmer. By exposing the Calva REPL to language models, it eliminates the gap between code generation and runtime verification. Developers can now ask an assistant to evaluate snippets, inspect live data structures, or perform structural edits while the code is actually running in their development environment.

At its core, the server offers a suite of high‑level tools that map directly to common Clojure development tasks. These include evaluating arbitrary code, creating or appending files with automatic bracket balancing, and performing structural edits like inserting or replacing top‑level forms. Each tool is designed to preserve the syntactic integrity of Clojure code, leveraging Parinfer for bracket balancing and built‑in linting to prevent accidental syntax errors. In addition, the server provides information‑retrieval tools: a symbol info lookup that returns docstrings and argument lists, and a clojuredocs.org lookup that surfaces examples and related symbols. These resources give the model contextual knowledge without having to fetch external documentation during a session.

For developers, this means that AI assistants can move beyond speculative code suggestions. They can run a function with real arguments, examine the resulting data structures, and then refine their suggestions based on actual output. This real‑time feedback loop is especially valuable for debugging, exploratory programming, and rapid prototyping. The server’s integration with VS Code through the Language Model API also allows developers to use CoPilot or any MCP‑compatible agent directly within their editor, keeping the workflow seamless.

Unique advantages of Calva Backseat Driver include its tight coupling to the Calva REPL, which is already a staple for Clojure developers in VS Code. The server respects editor settings that control line targeting and context padding, giving fine‑grained control over how much of the source is exposed to the model. Moreover, by exposing resources as both tools and experimental resources, it offers flexibility for agents that prefer a stateless or stateful interaction pattern.

In real‑world scenarios, teams can use the server to automate unit tests by having an assistant generate and evaluate test cases on the fly, or to refactor code bases with confidence that each edit is syntactically correct and immediately executable. The ability to append or replace top‑level forms with automatic formatting also streamlines large refactorings, ensuring that the assistant’s output remains clean and maintainable. Overall, Calva Backseat Driver empowers AI assistants to become true partners in the Clojure development cycle, bridging the gap between suggestion and execution.